Marc Baguelin 1,2. 1 Public Health England 2 London School of Hygiene & Tropical Medicine

Transcription

1 The cost-effectiveness of extending the seasonal influenza immunisation programme to school-aged children: the exemplar of the decision in the United Kingdom Marc Baguelin 1,2 1 Public Health England 2 London School of Hygiene & Tropical Medicine

2 Stopping the transmission of influenza and protecting the most vulnerable

3 Global Annual Incidence per 100,000 of severe influenza in children (meta analysis) Incidence (95%CI) Developing countries <6 mo. old 290( ) 0-11 mo. old 280 ( ) 0-59 mo. old 170 ( ) Industrialized countries <6 mo. old 340 ( ) 0 11 mo. old 230 ( ) 0 59 mo. old 120 (90-170) Nair et al., Lancet 2011; 378:

4 Current annual seasonal trivalent inactivated vaccine (TIV) programme in the UK All high risk groups under 65 years All 65+ year olds Problems : efficacy of TIV in elderly and the very young is poor most vulnerable groups are the elderly and the very young

5 And the situation in the UK Cromer D, et al., J Infect (2013)

6 What to do? the British philosophy Public health philosophy based on utilitarianism A government should use public resources to maximize the well being of the society Problem: how to quantify the well being when dealing with a public health intervention?

7 Quantifying the public health impact of an intervention Need a measurement which takes into account 1. The change in years of life 2. The change in term of quality of life (e.g. lost of mobility, pain ) The notion of Quality Adjusted Life Years (QALY) has been chosen as a measure in the UK One year of perfect health is equal to one QALY while death is 0 QALY Impact of intervention assessed using ICER measured in /QALY /QALY threshold for cost-effective interventions

8 Extensions to current programme Extend to low-risk: 2-4 years years 5-16 years 2-4 & years 2-16 years 2-16 & years 2-64 years Increasing cost Net additional cost 14m 282m Coverage assumed to be 50% in low-risk groups

9 Schematic of approach Epidemic parameters Reproduction number Incubation period Infectious period Susceptibility profile Mixing patterns Data RCGP Swabbing Serology Vaccine assumptions Coverage By age & risk By year & strain Efficacy Epidemic projections Outcomes Risk and age: CFR Hospit. QALY loss Costs Hospitalisation Vaccine Delivery Projections in relevant units, & CEA

10 Mathematical models of infectious diseases Compartmental models Based on the SEIR structure Include age structure and risk groups Easy to produce a model, difficult to fit to surveillance data in the case of influenza

11 Severity pyramid ILI different possible pathogens Only the top is observed by surveillance Dead Hospitalized Medically attended Knowledge fundamental for modelling H3N2 Symptomatic Infected

12 The problem of scale: nowcasting & forecasting using aggregate data Need to track depletion of susceptibles (i.e. infections) The data (i.e. # cases, or GP consultations, or deaths) gives us just a fraction of the infections What fraction? i.e. what fraction of infections are reported? Infections 40-80% 10-50% % % Clinical case Attend GP Recognised Reported Decreasing fraction Increasing time

13 Complex mathematical and statistical problem Evidence synthesis where mathematical modelling is used to linked different data sources Breaks down into elementary processes Bayesian approach to estimate uncertainty Dynamical transmission model and probabilistic observation model

14 Observation model Epidemiological model Vaccination model Transmission model Set of Contacts (A) POLYMOD survey for the UK Transmission matrix (β) Demographic data Transmissibility of the virus (q) Immunisation rate (ν) Vaccination uptake in UK Vaccination match Number of Infections (Z) Susceptibility profile (σ) Initial number of infections (l) Number of influenza cases in GP (m + ) Number of virologically positive (n + ) Ascertainment probability (ε) Outside infection (ψ) Number of virologically tested (n) Number of ILI cases (m)

15

16

17

18 Linking infection risks with disease risks Annual cases of infection by strain & age group Annual number of outcomes (e.g. hospit) Derive risk ratios Problems: Age groups not always identical assume outcomes distributed by pop size Some years very low (zero) infections estimated Labbase (burden) does not distinguish H1 and H3 Derived risk ratios from years when relatively large number of cases (and mostly 1 strain of flu A) A H1N1: 2000/1 and 2007/8 A H3N2: 2003/4, 2004/5 and 2006/7 B: 2000/1 and 2005/6

19 Costs & QALYs Costs of vaccination from DH NHS reference costs for treatment Non-fatal QALYs lost from literature Fatal cases assumed to lose average risk and age-specific life-expectancy adjusted for average age-specific QoL (from RCGP) Kind P et al. BMJ 1998;316:

20 Incremental cost-effectiveness Rank scenarios by increasing cost 2-4 years years 5-16 years 2-4 & years 2-4 & 5-16 years 2-4 & 5-16 & years 2-64 years Remove the dominated ones (more money, less QALYs saved Compute incremental CEAC (cost-effectiveness acceptability curve) Stop until not CE anymore

21 Increasing cost 20,000 /QALY threshold Increasing benefit

22 Are we vaccinating the kids to protect their grandparents? Ethical issue with recommending a treatment for the benefit of others

23 Death associated QALD incremental on current ILI associated QALD incremental on current

24 Cost-effectiveness of vaccinating children Cost-effective to vaccinate school children for their own benefit (i.e. ignoring benefits to other groups) ICER=8155 /QALY Mean Net Benefit 107m [90%CI: - 34m 390m] Burden exclusively in term of saved influenza episodes and hospitalisations

25 And finally the decision!

26 The future The implementation of the programme has started in 6 pilot areas this season for the 4-11, and nationally for the 2-3 It will be extended next year to more age groups

27 Main improvements / weaknesses Fitting: Serology Serial fitting (more swabbing data) Burden Deaths & life-expectancy (vaccinate everyone?) Linking infection & burden (different data sources) Low activity era: will this continue? Transmission Model suggests that children play a key role Cluster randomised trials (e.g. Loeb et al.) support this, but others have argued against

28 Acknowledgments Stefan Flasche, Anton Camacho, John Edmunds (London School of Hygiene and Tropical Medicine) Nikos Demiris (Athens University of Economics and Business) Liz Miller, Richard Pebody, Maria Zambon, Joanna Ellis, Albert-Jan Van Hoek, Julia Stowe, Pauline Kaye (Public Health England) Douglas Flemming (Royal College of Practitioner) Tom Barlow, Peter Grove (UK Department of Health)

29 Increment ICER ( /QALY) % < 20,000/QALY % < 30,000/QALY Net benefit in M 95% credibility interval Current 2-4 y (10.4 ; 247.0) 2-4 y 5-16 y (79.4 ; ) 5-16 y 2-16y (7.2 ; 208.2) y 2-16y &50-64y (-14.9 ; 327.9) y &50-64y 2-64y (-64.4 ; 919.9)